Date: Dec 5, 2012 1:11 AM
Author: Ray Koopman
Subject: Re: One other question about using Auq avg slope as a constant when<br> computing the other two regressions
On Dec 4, 9:18 pm, djh <halitsk...@att.net> wrote:

> The average slope ?Auq? of the regression c on (u,u^2)is the most

> stable of our five new average slopes; in particular:

>

> i) for method N and dicodon set 1, we get a perfect S/C split of the

> ?m?s? of the CI?s EVEN WHEN we plot the regressions of Auq on ALL

> singleton lengths for S and for C of all methods/sets/subsets, AS

> WELL AS when we do the same using the 12 length intervals instead of

> all available singleton lengths.

>

> ii) for method N and dicodon set 2, we get an almost perfect S/C split

> of the ?m?s? of the CI?s EVEN WHEN we plot the regressions of Auq on

> ALL singleton lengths for S and for C of all methods/sets/subsets, AS

> WELL AS when we do the same using the 12 length intervals instead of

> all available singleton lengths.

>

> And this ?stability? of Aug, coupled with my abysmal ignorance and

> naivete, leads me to ask the following question.

>

> When I user Ivor Welch?s module to compute the three new regessions

> Ruq, Rub, and Ruqb per singleton length interval, his module allows me

> to specify a constant which I now default to ?1?.

>

> So when I compute the regressions Rub and Ruq for a given singleton

> length L, would any possible benefit accrue from using the value of

> Auq for L (and, of course, the corresponding method, set, subset, and

> fold)?

>

> Or is this an entirely illegitimate way to use Auq as a ?constant?

> when computing Rub and Rubq?

>

> Thanks as always for considering this question.

What is the constant supposed to do or be?

How does it fit into the regression equation?